Provisioning Policies for Elastic Computing Environments

Resources experience dynamic load as demand fluctuates. Therefore, resource providers must estimate the appropriate amount of resources to purchase in order to meet variable user demand. With the relatively recent introduction of infrastructure-as-a-service (IaaS) clouds (e.g. Amazon EC2) resource providers may choose to outsource demand as needed. As a result, a resource provider may decide to decrease his initial capital outlay and purchase a smaller resource that meets the needs of his users the majority of the time while budgeting for future outsourcing costs. When bursts in demand exceed the capacity of the resource, a resource provider can use elastic computing to outsource excess demand to IaaS clouds based on a defined budget. To create efficient elastic environments, existing services must be extended with elastic computing functionality and resource provisioning policies that match resource deployments with demand must be developed. In this paper we consider an elastic environment that extends a local cluster resource with IaaS resources. We present resource provisioning policies to dynamically match resource supply with demand. Our policies balance the requirements of users and administrators, such as minimizing the monetary cost of the IaaS deployment and reducing job queued time. We develop a discrete event simulator, the elastic cloud simulator (ECS), to evaluate our policies using scientific workloads. Our results demonstrate that by outsourcing on a flexible basis instead of simply provisioning the maximum number of instances preemptively, we reduce the average queued time by up to 58% and cost by 38%. Our results also demonstrate that our multi-variable policies provide more flexibility in balancing budget and time requirements than typical single-variable reference policies, giving resource providers controls to manage their elastic environments.